Disclosure: This article contains an affiliate link to Retell AI. If you sign up through it, Automated Wealth Lab earns a commission at no extra cost to you. It never changes what we recommend — the affiliate only exists where the tool genuinely fits the workflow.
Every AI voice platform advertises a per-minute price: $0.05, $0.07, $0.14. None of them is the price of anything you actually buy. You don’t buy minutes. You buy resolved calls — the booking made, the question answered, the ticket deflected without a human. And the road from a minute sticker to a resolved call runs through stack fees, telephony, platform minimums, and the calls that fail — which you also pay for.
- The model (steal it, change the inputs)
- What a minute actually costs on each platform
- Retell: itemized, mid-pack, no platform fee
- Vapi: the cheapest sticker, the most invoices
- Bland: one bundled rate, a platform fee hiding behind it
- The model run: 1,000 calls × 4 minutes
- Costs the model doesn’t include (budget for them anyway)
- FAQ
- Methodology
- AWL verdict
- Sources
This benchmark builds one transparent model — 1,000 calls a month, 4 minutes average handled talk time — and prices Retell, Vapi and Bland through it, using each vendor’s published rates and pricing-calculator defaults as checked in October 2026. Every assumption is stated, so you can re-run it with your own call length and resolution rate.
The model (steal it, change the inputs)
- Volume: 1,000 calls/month.
- Average call length: 4.0 minutes of talk time → 4,000 billed minutes/month. (Appointment and support calls to small businesses cluster in the 2–6 minute range; 4 is the planning midpoint, and the sensitivity table below shows what changes if yours differ.)
- Resolved call: the agent completes the caller’s task end-to-end — booked, answered, resolved — with no human transfer and no callback needed.
- Resolution rates: modeled at 60%, 70% and 80%. These are scenarios, not measurements — no vendor publishes audited resolution rates, and anyone quoting you one without your call data is guessing. The spread matters more than the point estimate, as the results show.
Cost per resolved call = (monthly platform + usage cost ÷ 1,000 calls) ÷ resolution rate. The unresolved 20–40% of calls aren’t free — they’re the failure tax, and they’re already in the formula.
What a minute actually costs on each platform
Retell: itemized, mid-pack, no platform fee
Retell bills the stack in the open: $0.055/min voice infrastructure + voice from $0.015/min + your LLM (from ~$0.002/min for the cheapest models to $0.064/min for its recommended frontier options) + telephony at $0.015/min on a US Retell number, or $0 if you bring your own SIP trunk. Headline range: $0.07–0.31/min. Retell’s own calculator default lands at $0.11/min before telephony — call it $0.125/min with a US number, the configuration we model. No platform fee; a phone number is $2/month; 20 concurrent calls included (extra lines $8/month each); $10 free credits to start.
Two fine-print items worth knowing: prompts over ~4,000 tokens scale the billed duration of the whole call, so a bloated system prompt taxes every minute; and burst calls above your concurrency limit attract a surcharge on the entire call.
Vapi: the cheapest sticker, the most invoices
Vapi charges $0.05/min for hosting — and that’s the only line Vapi controls. Speech-to-text, the LLM and the voice are passed through at provider cost, no markup: Vapi’s own calculator, using Deepgram transcription (~$0.0095/min), an OpenAI model ($0.0077–0.0452/min) and an ElevenLabs voice ($0.0146–0.0238/min), totals $0.082–0.129/min at 1,000 minutes a month, with transport free on Vapi’s own numbers (Twilio inbound adds $0.008/min if you bring your carrier). We model the midpoint, $0.105/min.
The platform-fee trap is at the plan level: pay-as-you-go includes only 4 concurrent calls (extra lines $10/month each). The Core package — $29/month — raises that to 10 lines and 5 numbers. Pro is priced at 10% of hosting spend with a $999/month minimum: a serious cliff that low-volume users never need to touch. And if you need HIPAA, Vapi lists it as a $2,000/month add-on — for a healthcare use case, that single line can exceed the entire call bill.
Bland: one bundled rate, a platform fee hiding behind it
Bland bundles LLM, speech-to-text and text-to-speech into a single rate — no token meter running underneath. Start: $0.14/min, no platform fee, 10 concurrent calls, capped at 100 calls/day. Build: $0.12/min plus a $299/month platform fee, 50 concurrent calls. A Scale tier ($0.11/min + $499/month) appears in Bland’s billing docs. Telephony is not in the rate — carrier fees apply separately — and warm transfers to a human are billed per minute ($0.05/min on Start, $0.04 on Build). Note that these rates already moved once recently: Start was $0.09/min until December 2025.
The model run: 1,000 calls × 4 minutes
| Platform (modeled config) | All-in $/min | Monthly usage | Fixed fees | Cost per call | Cost per resolved call @70% |
|---|---|---|---|---|---|
| Vapi (pay-as-you-go) | $0.105 | $420 | $0 | $0.42 | $0.60 |
| Vapi (Core, $29/mo) | $0.105 | $420 | $29 | $0.45 | $0.64 |
| Retell (default build + US number) | $0.125 | $500 | $2 | $0.50 | $0.72 |
| Bland Start | $0.14 | $560 | $0 (+ telephony) | $0.56 | $0.80 |
| Bland Build | $0.12 | $480 | $299 | $0.78 | $1.11 |
Three findings fall out:
1. At 1,000 calls a month, Bland Build is a trap. The $299 platform fee buys a $0.02/min discount worth $80 at this volume. Build only breaks even against Start at 299 ÷ 0.02 = 14,950 minutes/month — roughly 3,700 four-minute calls. Below that, Start wins inside Bland’s own lineup.
2. Vapi’s sticker survives contact with the model — barely. The $0.05 headline becomes $0.105 assembled, still the cheapest per call here, if you stay on pay-as-you-go limits and don’t need HIPAA or concurrency. Its cost advantage is ~$0.08–0.14 per resolved call over Retell at this volume — real money at 10,000 calls, minor at 1,000.
3. The platform choice matters less than the resolution rate. Sensitivity, cost per resolved call:
| Platform | @60% resolved | @70% resolved | @80% resolved |
|---|---|---|---|
| Vapi (PAYG) | $0.70 | $0.60 | $0.53 |
| Retell | $0.84 | $0.72 | $0.63 |
| Bland Start | $0.93 | $0.80 | $0.70 |
Moving resolution from 60% to 80% — better prompts, better knowledge base, tighter call scope — cuts cost per resolution by ~25% on every platform. Switching from the most expensive to the cheapest platform at a fixed resolution saves ~10–25%. Operations beats shopping. This is the same second-meter pattern we documented for agent platforms in The Second Meter: the meter you don’t watch costs more than the one on the pricing page.
Costs the model doesn’t include (budget for them anyway)
- Failed and abandoned calls. Every platform bills for connected minutes whether or not anything got resolved. They’re in the resolution-rate math above, but a high hangup rate in the first 10 seconds still burns real minutes across 1,000 calls.
- Concurrency spikes. Retell includes 20 lines, then $8/line/month; Vapi pay-as-you-go includes 4, then $10/line/month; Bland Start caps at 10 concurrent and 100 calls/day. A Monday-morning spike is a plan event, not a rounding error.
- Transfers. On Bland, warm-transfer minutes are billed ($0.05/min on Start) — a call the agent “handles” by handing off still meters twice.
- Compliance. Vapi’s HIPAA add-on ($2,000/month) vs Retell’s self-serve BAA at no extra fee, per current comparison trackers — a $24,000/year difference that swamps every per-minute gap in this article for a medical or dental use case. (We priced exactly that scenario in AI receptionist as a micro-SaaS.)
- Your build time. All three are infrastructure, not receptionists. The agent’s prompt, tools and calendar plumbing are yours to build — cheapest on Retell’s low-code builder, most flexible on Vapi’s API, most managed on Bland.
FAQ
What is a good cost per resolved call for an AI voice agent?
From this model, $0.50–$0.95 per resolved call at 1,000 calls/month is the realistic 2026 band for small-business workloads, depending on platform and resolution rate. A human-handled simple call typically costs several dollars in labor time, which is why the economics work even at the expensive end — but only if the resolution rate is real. Measure yours; don’t borrow the model’s 70%.
Is Vapi really cheaper than Retell?
Per assembled minute, usually by a little: Vapi’s own calculator puts its default stack at $0.082–0.129/min; Retell’s default build is ~$0.125/min with telephony. The gap can invert — a pricey voice or model choice on Vapi, or free BYO-SIP telephony on Retell, flips it. And plan-level items (Vapi’s $999-minimum Pro, $2,000 HIPAA add-on; Retell’s included concurrency) matter more than the per-minute delta at small volumes.
Why does Bland cost more in this model if its rate is “all-in”?
Because at 1,000 calls the all-in rate ($0.14 on Start) is higher than the assembled competition, and the discounted Build rate ($0.12) carries a $299/month fee that needs ~14,950 minutes a month to pay for itself. Bland’s bundle buys predictability — no token surprises — and at high volumes its Scale rate ($0.11) becomes competitive. Volume decides.
Do these platforms charge for calls the agent fails to resolve?
Yes. All three bill connected minutes regardless of outcome. That’s why cost per resolved call — not per minute, not per call — is the only honest unit: it prices the failures in.
Which platform should a solo founder pick for a first agent?
If you’re testing an idea at low volume: Retell’s pay-as-you-go (no platform fee, $10 free credits, 20 concurrent calls, low-code builder) is the lowest-risk start — Retell is the link we use ourselves for that reason. If you’re a developer who wants to choose every model and already holds provider accounts, Vapi’s pass-through pricing rewards you. If you need one predictable invoice for a finance team, Bland Start — until your minutes approach 15,000 a month.
How do I raise my resolution rate?
Narrow the agent’s job (one call type per agent), ground it in a real knowledge base instead of a long prompt (which also avoids Retell’s prompt-size billing quirk), connect the calendar/CRM tools it actually needs, and review failed-call transcripts weekly. Every point of resolution rate is worth more than any platform switch in this article.
Methodology
What we did: Took each vendor’s published component rates and official pricing-calculator defaults (Retell, Vapi) and published plan rates (Bland) as checked by pricing trackers in September–October 2026, then ran one fixed model — 1,000 calls, 4.0 minutes average, resolution scenarios of 60/70/80% — through each platform’s billing rules. All arithmetic is shown and re-runnable.
What we did NOT do: We did not place 1,000 calls, measure any platform’s actual resolution rate, or test voice quality, latency or reliability. The resolution rates are explicitly labeled scenarios, not vendor claims or measurements. Per-minute “all-in” figures are calculator defaults and mid-range configurations; your model/voice/telephony choices will move them, which is why the ranges are published in the text.
Conflict disclosure: Retell AI is an affiliate partner (15% commission for 2 years, per its published affiliate program page, checked October 2026). That did not change the model, in which Retell finishes mid-pack on cost — the article says so, with the arithmetic shown. Prices checked October 2026.
AWL verdict
Price per resolved call, not per minute, and the ranking at small-business volume is: Vapi cheapest if you’re technical and your needs fit its pay-as-you-go limits ($0.60/resolved in our model); Retell the balanced default — itemized pricing, no platform fee, enough included concurrency, and no compliance ambush ($0.72); Bland the predictability pick whose discounted tier is a trap below ~15,000 minutes a month ($0.80 on Start, $1.11 on Build at this volume). But the verdict under the verdict: your resolution rate is a bigger cost lever than your platform. Fix the agent before you shop for a cheaper meter.
Sources
- Retell AI pricing breakdown — component rates ($0.055 infrastructure, voice from $0.015, telephony $0.015), $0.07–0.31/min range, $2/mo number, 20 concurrent calls, prompt-size billing rule (checked 13 October 2026): https://imisofts.com/blog/retell-ai-pricing/
- Retell AI pricing review — calculator default ≈ $0.11/min, $10 free credits, concurrency add-ons (checked 24 September 2026): https://www.getmacha.com/blog/retell-ai-complete-guide
- Vapi pricing analysis — $0.05/min hosting, pass-through component ranges, calculator total $0.082–0.129/min, Core $29/mo, Pro $999/mo minimum, HIPAA $2,000/mo (checked 24 September 2026): https://www.getmacha.com/blog/vapi-ai-pricing-explained
- Vapi pricing plans and concurrency limits, via CloudTalk, September 2026: https://www.cloudtalk.io/blog/vapi-ai-pricing/
- Bland AI pricing — Start $0.14/min, Build $0.12/min + $299/mo, Scale $0.11/min + $499/mo in billing docs, transfer-minute rates, December 2025 rate change, via Macha, October 2026: https://www.getmacha.com/blog/bland-ai-pricing-explained
- Bland AI plan structure and bundled-rate composition, via Bland’s own pricing explainer, October 2026: https://www.bland.ai/blog/how-much-does-a-voice-ai-platform-cost
- Cross-platform per-minute comparison (Retell defaults, Bland plan rates), via Cartesia, October 2026: https://www.cartesia.ai/learn/ai-voice-agent-platforms
- Retell vs Vapi plan details — concurrency, BAA/HIPAA treatment (October 2026): https://www.snapsdr.com/blog/vapi-alternatives-2026
